Few-shot Learning with Class-Covariance Metric for Hyperspectral Image Classification, TIP, 2022.

Bobo Xi, Jiaojiao Li, Yunsong Li, Rui song, Danfeng Hong and Jocelyn Chanussot.

---

Code for the paper: Few-shot Learning with Class-Covariance Metric for Hyperspectral Image Classification.

Fig. 1: The architecture of the proposed CMFSL for HSIC. Based on the class-covariance metric, the classification process is completed by the episode-based collaboratively meta-training of the source and target data sets, and the episode-based meta-test of the target data set. Notably, the embedding feature extractor comprises a new SPRM and a novel LXConvNet.

Training and Test Process

1. Please prepare the training and test data as operated in the paper. And the websites to access the datasets are also provided. The used OCBS band selection method is referred to https://github.com/tanmlh
2. Run "trainMetaDataProcess.py" to generate the meta-training data
3. Run the 'CMFSL_UP_main.py' to reproduce the CMFSL results on Pavia University data set.

We have successfully tested it on Ubuntu 16.04 with PyTorch 1.1.0. Below is the classification map with five shots of training samples from each class.

Fig. 2: The composite false-color image, groundtruth, and classification map of Pavia University dataset.

References

If you find this code helpful, please kindly cite:

[1] B. Xi, J. Li, Y. Li, R. Song, D. Hong and J. Chanussot, "Few-Shot Learning With Class-Covariance Metric for Hyperspectral Image Classification," in IEEE Transactions on Image Processing, vol. 31, pp. 5079-5092, 2022, doi: 10.1109/TIP.2022.3192712.

Citation Details

BibTeX entry:

@ARTICLE{Xi_2022TIP_CMFSL,
  author={Xi, Bobo and Li, Jiaojiao and Li, Yunsong and Song, Rui and Hong, Danfeng and Chanussot, Jocelyn},
  journal={IEEE Transactions on Image Processing}, 
  title={Few-Shot Learning With Class-Covariance Metric for Hyperspectral Image Classification}, 
  year={2022},
  volume={31},
  number={},
  pages={5079-5092},
  doi={10.1109/TIP.2022.3192712}}

Licensing

Copyright (C) 2022 Bobo Xi

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 3 of the License.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program.